Horses 2010
Tue Oct 23 14:29:20 2018
1 Overview
The raster GIS layer on Horses (2010) described in this document are part of the Gridded Livestock of the World database (GLW v3.1).
The GLW database was developped to meet three objectives:
- Collect, harmonize and disseminate sub-national global data on livestock distribution;
- Predict livestock numbers in areas with missing informations on livestock census counts;
- Provide an statistically-informed estimate of how livestock may distribute within census unit through downscaling algoritms.
Two main methods have been described in the litterature for distributing census data at the pixel level. The areal-weighted (AW) method simply considers that all pixels of the census area are equally suitable, and assign them an equal weight. It simply corresponds to the density of animals per km2 in the census unit multiplied by the pixel area. The dasymetric method (DA) assigns different weights to different pixels based on high resolution environmental predictor variables and Random Forest models, and the animal census counts are distributed according to these weights.
The AW and DA GLW global distribution data fit different purposes. The DA GLW models provide an estimate of how livestock species may be distributed within census areas. However, spatial predictors (e.g. human population density, vegetation indices, topography, etc.) that are used to derived the downscaling weights may introduce some uncontrolled counfonding effects for users willing to quantify the effect of livestock alongside these spatial predictors on an outcome. Similarly, the DA models may introduce circularity for users willing to use livestock data to study their impact on some these spatial factors, such as land-use, for example. The AW GLW models are free of any influence of spatial predictor variables, but produces more unrealistic distribution patterns, especially in large census areas containing a wide range of different land-use and farming conditions.
This report presents snapshots and links to high resolution maps of the different data layers available with the GLW v3.1 data package for Horses (2010). Three GIS layers are provided, as geotiff, with an extent of X:-180 to 180 / Y:-90 to 90, a resolution of 0.083333 decimal degrees, and with the projection CRS “+proj=longlat”:
- DA GLW model (5_Ho_2010_Da.tif): This layer contains the DA animal numbers per pixel, with weight estimated by the Random Forest model. Note that the totals number of animals per census polygon of the DA and AW GLW models should be identical.
- AW GLW model (6_Ho_2010_Aw.tif): This layer contains the AW animal numbers per pixel in polygons where data where available in the subnational census, and AW predictions from the the model in census area where the information was missing.
- AW GLW prediction status (7_Ho_2010_Ps.tif): This layer contains a value of 0 in pixels where the AW animal number is from a census, and value of 1 if the value was predicted by the GLW model.
2 Snapshots
Note that the snapshot below were produced in the equal-area pseudocylindrical Eckert IV projection for a proper representation of densities. The standard lat/long visualisation of global raster data tend to visually over-represent animal densities in pixels located in northern latitudes as they cover a much lower surface on earth than those close to the equator. Altough the data files are distributed in lat/long, we recommend the use of an equal-area projection for displaying our livestock density data.
2.3 Areas predicted in the AW model (7_Ho_2010_Ps.tif)
A high-resolution version can be visualised here.
3 Description of sub-national census data
3.1 Average Spatial Resolution (ASR)
The average spatial resolution (ASR) gives an estimate of the average resolution of each spatial unit used to build the GLW models. It is estimated, for each country as \(\sqrt[root]{Area/N_u}\), where Area is the surface of the country in square km, and \(N_u\) is the number of sub-national units per country.
3.2 Year of the sub-national input data
The following histogram and maps show the year of sub-national census data used for the GLW models.
